1 // SPDX-License-Identifier: GPL-2.0+
3 * User-space Probes (UProbes)
5 * Copyright (C) IBM Corporation, 2008-2012
9 * Copyright (C) 2011-2012 Red Hat, Inc., Peter Zijlstra
12 #include <linux/kernel.h>
13 #include <linux/highmem.h>
14 #include <linux/pagemap.h> /* read_mapping_page */
15 #include <linux/slab.h>
16 #include <linux/sched.h>
17 #include <linux/sched/mm.h>
18 #include <linux/sched/coredump.h>
19 #include <linux/export.h>
20 #include <linux/rmap.h> /* anon_vma_prepare */
21 #include <linux/mmu_notifier.h> /* set_pte_at_notify */
22 #include <linux/swap.h> /* try_to_free_swap */
23 #include <linux/ptrace.h> /* user_enable_single_step */
24 #include <linux/kdebug.h> /* notifier mechanism */
25 #include "../../mm/internal.h" /* munlock_vma_page */
26 #include <linux/percpu-rwsem.h>
27 #include <linux/task_work.h>
28 #include <linux/shmem_fs.h>
29 #include <linux/khugepaged.h>
31 #include <linux/uprobes.h>
33 #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES)
34 #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE
36 static struct rb_root uprobes_tree = RB_ROOT;
38 * allows us to skip the uprobe_mmap if there are no uprobe events active
39 * at this time. Probably a fine grained per inode count is better?
41 #define no_uprobe_events() RB_EMPTY_ROOT(&uprobes_tree)
43 static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */
45 #define UPROBES_HASH_SZ 13
46 /* serialize uprobe->pending_list */
47 static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ];
48 #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
50 DEFINE_STATIC_PERCPU_RWSEM(dup_mmap_sem);
52 /* Have a copy of original instruction */
53 #define UPROBE_COPY_INSN 0
56 struct rb_node rb_node; /* node in the rb tree */
58 struct rw_semaphore register_rwsem;
59 struct rw_semaphore consumer_rwsem;
60 struct list_head pending_list;
61 struct uprobe_consumer *consumers;
62 struct inode *inode; /* Also hold a ref to inode */
64 loff_t ref_ctr_offset;
68 * The generic code assumes that it has two members of unknown type
69 * owned by the arch-specific code:
71 * insn - copy_insn() saves the original instruction here for
72 * arch_uprobe_analyze_insn().
74 * ixol - potentially modified instruction to execute out of
75 * line, copied to xol_area by xol_get_insn_slot().
77 struct arch_uprobe arch;
80 struct delayed_uprobe {
81 struct list_head list;
82 struct uprobe *uprobe;
86 static DEFINE_MUTEX(delayed_uprobe_lock);
87 static LIST_HEAD(delayed_uprobe_list);
90 * Execute out of line area: anonymous executable mapping installed
91 * by the probed task to execute the copy of the original instruction
92 * mangled by set_swbp().
94 * On a breakpoint hit, thread contests for a slot. It frees the
95 * slot after singlestep. Currently a fixed number of slots are
99 wait_queue_head_t wq; /* if all slots are busy */
100 atomic_t slot_count; /* number of in-use slots */
101 unsigned long *bitmap; /* 0 = free slot */
103 struct vm_special_mapping xol_mapping;
104 struct page *pages[2];
106 * We keep the vma's vm_start rather than a pointer to the vma
107 * itself. The probed process or a naughty kernel module could make
108 * the vma go away, and we must handle that reasonably gracefully.
110 unsigned long vaddr; /* Page(s) of instruction slots */
114 * valid_vma: Verify if the specified vma is an executable vma
115 * Relax restrictions while unregistering: vm_flags might have
116 * changed after breakpoint was inserted.
117 * - is_register: indicates if we are in register context.
118 * - Return 1 if the specified virtual address is in an
121 static bool valid_vma(struct vm_area_struct *vma, bool is_register)
123 vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_MAYSHARE;
128 return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC;
131 static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
133 return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
136 static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
138 return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
142 * __replace_page - replace page in vma by new page.
143 * based on replace_page in mm/ksm.c
145 * @vma: vma that holds the pte pointing to page
146 * @addr: address the old @page is mapped at
147 * @old_page: the page we are replacing by new_page
148 * @new_page: the modified page we replace page by
150 * If @new_page is NULL, only unmap @old_page.
152 * Returns 0 on success, negative error code otherwise.
154 static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
155 struct page *old_page, struct page *new_page)
157 struct mm_struct *mm = vma->vm_mm;
158 struct page_vma_mapped_walk pvmw = {
159 .page = compound_head(old_page),
164 struct mmu_notifier_range range;
166 mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, mm, addr,
170 err = mem_cgroup_charge(page_folio(new_page), vma->vm_mm,
176 /* For try_to_free_swap() below */
179 mmu_notifier_invalidate_range_start(&range);
181 if (!page_vma_mapped_walk(&pvmw))
183 VM_BUG_ON_PAGE(addr != pvmw.address, old_page);
187 page_add_new_anon_rmap(new_page, vma, addr, false);
188 lru_cache_add_inactive_or_unevictable(new_page, vma);
190 /* no new page, just dec_mm_counter for old_page */
191 dec_mm_counter(mm, MM_ANONPAGES);
193 if (!PageAnon(old_page)) {
194 dec_mm_counter(mm, mm_counter_file(old_page));
195 inc_mm_counter(mm, MM_ANONPAGES);
198 flush_cache_page(vma, addr, pte_pfn(*pvmw.pte));
199 ptep_clear_flush_notify(vma, addr, pvmw.pte);
201 set_pte_at_notify(mm, addr, pvmw.pte,
202 mk_pte(new_page, vma->vm_page_prot));
204 page_remove_rmap(old_page, vma, false);
205 if (!page_mapped(old_page))
206 try_to_free_swap(old_page);
207 page_vma_mapped_walk_done(&pvmw);
212 mmu_notifier_invalidate_range_end(&range);
213 unlock_page(old_page);
218 * is_swbp_insn - check if instruction is breakpoint instruction.
219 * @insn: instruction to be checked.
220 * Default implementation of is_swbp_insn
221 * Returns true if @insn is a breakpoint instruction.
223 bool __weak is_swbp_insn(uprobe_opcode_t *insn)
225 return *insn == UPROBE_SWBP_INSN;
229 * is_trap_insn - check if instruction is breakpoint instruction.
230 * @insn: instruction to be checked.
231 * Default implementation of is_trap_insn
232 * Returns true if @insn is a breakpoint instruction.
234 * This function is needed for the case where an architecture has multiple
235 * trap instructions (like powerpc).
237 bool __weak is_trap_insn(uprobe_opcode_t *insn)
239 return is_swbp_insn(insn);
242 static void copy_from_page(struct page *page, unsigned long vaddr, void *dst, int len)
244 void *kaddr = kmap_atomic(page);
245 memcpy(dst, kaddr + (vaddr & ~PAGE_MASK), len);
246 kunmap_atomic(kaddr);
249 static void copy_to_page(struct page *page, unsigned long vaddr, const void *src, int len)
251 void *kaddr = kmap_atomic(page);
252 memcpy(kaddr + (vaddr & ~PAGE_MASK), src, len);
253 kunmap_atomic(kaddr);
256 static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode)
258 uprobe_opcode_t old_opcode;
262 * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here.
263 * We do not check if it is any other 'trap variant' which could
264 * be conditional trap instruction such as the one powerpc supports.
266 * The logic is that we do not care if the underlying instruction
267 * is a trap variant; uprobes always wins over any other (gdb)
270 copy_from_page(page, vaddr, &old_opcode, UPROBE_SWBP_INSN_SIZE);
271 is_swbp = is_swbp_insn(&old_opcode);
273 if (is_swbp_insn(new_opcode)) {
274 if (is_swbp) /* register: already installed? */
277 if (!is_swbp) /* unregister: was it changed by us? */
284 static struct delayed_uprobe *
285 delayed_uprobe_check(struct uprobe *uprobe, struct mm_struct *mm)
287 struct delayed_uprobe *du;
289 list_for_each_entry(du, &delayed_uprobe_list, list)
290 if (du->uprobe == uprobe && du->mm == mm)
295 static int delayed_uprobe_add(struct uprobe *uprobe, struct mm_struct *mm)
297 struct delayed_uprobe *du;
299 if (delayed_uprobe_check(uprobe, mm))
302 du = kzalloc(sizeof(*du), GFP_KERNEL);
308 list_add(&du->list, &delayed_uprobe_list);
312 static void delayed_uprobe_delete(struct delayed_uprobe *du)
320 static void delayed_uprobe_remove(struct uprobe *uprobe, struct mm_struct *mm)
322 struct list_head *pos, *q;
323 struct delayed_uprobe *du;
328 list_for_each_safe(pos, q, &delayed_uprobe_list) {
329 du = list_entry(pos, struct delayed_uprobe, list);
331 if (uprobe && du->uprobe != uprobe)
333 if (mm && du->mm != mm)
336 delayed_uprobe_delete(du);
340 static bool valid_ref_ctr_vma(struct uprobe *uprobe,
341 struct vm_area_struct *vma)
343 unsigned long vaddr = offset_to_vaddr(vma, uprobe->ref_ctr_offset);
345 return uprobe->ref_ctr_offset &&
347 file_inode(vma->vm_file) == uprobe->inode &&
348 (vma->vm_flags & (VM_WRITE|VM_SHARED)) == VM_WRITE &&
349 vma->vm_start <= vaddr &&
353 static struct vm_area_struct *
354 find_ref_ctr_vma(struct uprobe *uprobe, struct mm_struct *mm)
356 struct vm_area_struct *tmp;
358 for (tmp = mm->mmap; tmp; tmp = tmp->vm_next)
359 if (valid_ref_ctr_vma(uprobe, tmp))
366 __update_ref_ctr(struct mm_struct *mm, unsigned long vaddr, short d)
370 struct vm_area_struct *vma;
377 ret = get_user_pages_remote(mm, vaddr, 1,
378 FOLL_WRITE, &page, &vma, NULL);
379 if (unlikely(ret <= 0)) {
381 * We are asking for 1 page. If get_user_pages_remote() fails,
382 * it may return 0, in that case we have to return error.
384 return ret == 0 ? -EBUSY : ret;
387 kaddr = kmap_atomic(page);
388 ptr = kaddr + (vaddr & ~PAGE_MASK);
390 if (unlikely(*ptr + d < 0)) {
391 pr_warn("ref_ctr going negative. vaddr: 0x%lx, "
392 "curr val: %d, delta: %d\n", vaddr, *ptr, d);
400 kunmap_atomic(kaddr);
405 static void update_ref_ctr_warn(struct uprobe *uprobe,
406 struct mm_struct *mm, short d)
408 pr_warn("ref_ctr %s failed for inode: 0x%lx offset: "
409 "0x%llx ref_ctr_offset: 0x%llx of mm: 0x%pK\n",
410 d > 0 ? "increment" : "decrement", uprobe->inode->i_ino,
411 (unsigned long long) uprobe->offset,
412 (unsigned long long) uprobe->ref_ctr_offset, mm);
415 static int update_ref_ctr(struct uprobe *uprobe, struct mm_struct *mm,
418 struct vm_area_struct *rc_vma;
419 unsigned long rc_vaddr;
422 rc_vma = find_ref_ctr_vma(uprobe, mm);
425 rc_vaddr = offset_to_vaddr(rc_vma, uprobe->ref_ctr_offset);
426 ret = __update_ref_ctr(mm, rc_vaddr, d);
428 update_ref_ctr_warn(uprobe, mm, d);
434 mutex_lock(&delayed_uprobe_lock);
436 ret = delayed_uprobe_add(uprobe, mm);
438 delayed_uprobe_remove(uprobe, mm);
439 mutex_unlock(&delayed_uprobe_lock);
446 * Expect the breakpoint instruction to be the smallest size instruction for
447 * the architecture. If an arch has variable length instruction and the
448 * breakpoint instruction is not of the smallest length instruction
449 * supported by that architecture then we need to modify is_trap_at_addr and
450 * uprobe_write_opcode accordingly. This would never be a problem for archs
451 * that have fixed length instructions.
453 * uprobe_write_opcode - write the opcode at a given virtual address.
454 * @auprobe: arch specific probepoint information.
455 * @mm: the probed process address space.
456 * @vaddr: the virtual address to store the opcode.
457 * @opcode: opcode to be written at @vaddr.
459 * Called with mm->mmap_lock held for write.
460 * Return 0 (success) or a negative errno.
462 int uprobe_write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
463 unsigned long vaddr, uprobe_opcode_t opcode)
465 struct uprobe *uprobe;
466 struct page *old_page, *new_page;
467 struct vm_area_struct *vma;
468 int ret, is_register, ref_ctr_updated = 0;
469 bool orig_page_huge = false;
470 unsigned int gup_flags = FOLL_FORCE;
472 is_register = is_swbp_insn(&opcode);
473 uprobe = container_of(auprobe, struct uprobe, arch);
477 gup_flags |= FOLL_SPLIT_PMD;
478 /* Read the page with vaddr into memory */
479 ret = get_user_pages_remote(mm, vaddr, 1, gup_flags,
480 &old_page, &vma, NULL);
484 ret = verify_opcode(old_page, vaddr, &opcode);
488 if (WARN(!is_register && PageCompound(old_page),
489 "uprobe unregister should never work on compound page\n")) {
494 /* We are going to replace instruction, update ref_ctr. */
495 if (!ref_ctr_updated && uprobe->ref_ctr_offset) {
496 ret = update_ref_ctr(uprobe, mm, is_register ? 1 : -1);
504 if (!is_register && !PageAnon(old_page))
507 ret = anon_vma_prepare(vma);
512 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
516 __SetPageUptodate(new_page);
517 copy_highpage(new_page, old_page);
518 copy_to_page(new_page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
521 struct page *orig_page;
524 VM_BUG_ON_PAGE(!PageAnon(old_page), old_page);
526 index = vaddr_to_offset(vma, vaddr & PAGE_MASK) >> PAGE_SHIFT;
527 orig_page = find_get_page(vma->vm_file->f_inode->i_mapping,
531 if (PageUptodate(orig_page) &&
532 pages_identical(new_page, orig_page)) {
533 /* let go new_page */
537 if (PageCompound(orig_page))
538 orig_page_huge = true;
544 ret = __replace_page(vma, vaddr, old_page, new_page);
550 if (unlikely(ret == -EAGAIN))
553 /* Revert back reference counter if instruction update failed. */
554 if (ret && is_register && ref_ctr_updated)
555 update_ref_ctr(uprobe, mm, -1);
557 /* try collapse pmd for compound page */
558 if (!ret && orig_page_huge)
559 collapse_pte_mapped_thp(mm, vaddr);
565 * set_swbp - store breakpoint at a given address.
566 * @auprobe: arch specific probepoint information.
567 * @mm: the probed process address space.
568 * @vaddr: the virtual address to insert the opcode.
570 * For mm @mm, store the breakpoint instruction at @vaddr.
571 * Return 0 (success) or a negative errno.
573 int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
575 return uprobe_write_opcode(auprobe, mm, vaddr, UPROBE_SWBP_INSN);
579 * set_orig_insn - Restore the original instruction.
580 * @mm: the probed process address space.
581 * @auprobe: arch specific probepoint information.
582 * @vaddr: the virtual address to insert the opcode.
584 * For mm @mm, restore the original opcode (opcode) at @vaddr.
585 * Return 0 (success) or a negative errno.
588 set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
590 return uprobe_write_opcode(auprobe, mm, vaddr,
591 *(uprobe_opcode_t *)&auprobe->insn);
594 static struct uprobe *get_uprobe(struct uprobe *uprobe)
596 refcount_inc(&uprobe->ref);
600 static void put_uprobe(struct uprobe *uprobe)
602 if (refcount_dec_and_test(&uprobe->ref)) {
604 * If application munmap(exec_vma) before uprobe_unregister()
605 * gets called, we don't get a chance to remove uprobe from
606 * delayed_uprobe_list from remove_breakpoint(). Do it here.
608 mutex_lock(&delayed_uprobe_lock);
609 delayed_uprobe_remove(uprobe, NULL);
610 mutex_unlock(&delayed_uprobe_lock);
615 static __always_inline
616 int uprobe_cmp(const struct inode *l_inode, const loff_t l_offset,
617 const struct uprobe *r)
619 if (l_inode < r->inode)
622 if (l_inode > r->inode)
625 if (l_offset < r->offset)
628 if (l_offset > r->offset)
634 #define __node_2_uprobe(node) \
635 rb_entry((node), struct uprobe, rb_node)
637 struct __uprobe_key {
642 static inline int __uprobe_cmp_key(const void *key, const struct rb_node *b)
644 const struct __uprobe_key *a = key;
645 return uprobe_cmp(a->inode, a->offset, __node_2_uprobe(b));
648 static inline int __uprobe_cmp(struct rb_node *a, const struct rb_node *b)
650 struct uprobe *u = __node_2_uprobe(a);
651 return uprobe_cmp(u->inode, u->offset, __node_2_uprobe(b));
654 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
656 struct __uprobe_key key = {
660 struct rb_node *node = rb_find(&key, &uprobes_tree, __uprobe_cmp_key);
663 return get_uprobe(__node_2_uprobe(node));
669 * Find a uprobe corresponding to a given inode:offset
670 * Acquires uprobes_treelock
672 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
674 struct uprobe *uprobe;
676 spin_lock(&uprobes_treelock);
677 uprobe = __find_uprobe(inode, offset);
678 spin_unlock(&uprobes_treelock);
683 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
685 struct rb_node *node;
687 node = rb_find_add(&uprobe->rb_node, &uprobes_tree, __uprobe_cmp);
689 return get_uprobe(__node_2_uprobe(node));
691 /* get access + creation ref */
692 refcount_set(&uprobe->ref, 2);
697 * Acquire uprobes_treelock.
698 * Matching uprobe already exists in rbtree;
699 * increment (access refcount) and return the matching uprobe.
701 * No matching uprobe; insert the uprobe in rb_tree;
702 * get a double refcount (access + creation) and return NULL.
704 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
708 spin_lock(&uprobes_treelock);
709 u = __insert_uprobe(uprobe);
710 spin_unlock(&uprobes_treelock);
716 ref_ctr_mismatch_warn(struct uprobe *cur_uprobe, struct uprobe *uprobe)
718 pr_warn("ref_ctr_offset mismatch. inode: 0x%lx offset: 0x%llx "
719 "ref_ctr_offset(old): 0x%llx ref_ctr_offset(new): 0x%llx\n",
720 uprobe->inode->i_ino, (unsigned long long) uprobe->offset,
721 (unsigned long long) cur_uprobe->ref_ctr_offset,
722 (unsigned long long) uprobe->ref_ctr_offset);
725 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset,
726 loff_t ref_ctr_offset)
728 struct uprobe *uprobe, *cur_uprobe;
730 uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
734 uprobe->inode = inode;
735 uprobe->offset = offset;
736 uprobe->ref_ctr_offset = ref_ctr_offset;
737 init_rwsem(&uprobe->register_rwsem);
738 init_rwsem(&uprobe->consumer_rwsem);
740 /* add to uprobes_tree, sorted on inode:offset */
741 cur_uprobe = insert_uprobe(uprobe);
742 /* a uprobe exists for this inode:offset combination */
744 if (cur_uprobe->ref_ctr_offset != uprobe->ref_ctr_offset) {
745 ref_ctr_mismatch_warn(cur_uprobe, uprobe);
746 put_uprobe(cur_uprobe);
748 return ERR_PTR(-EINVAL);
757 static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
759 down_write(&uprobe->consumer_rwsem);
760 uc->next = uprobe->consumers;
761 uprobe->consumers = uc;
762 up_write(&uprobe->consumer_rwsem);
766 * For uprobe @uprobe, delete the consumer @uc.
767 * Return true if the @uc is deleted successfully
770 static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
772 struct uprobe_consumer **con;
775 down_write(&uprobe->consumer_rwsem);
776 for (con = &uprobe->consumers; *con; con = &(*con)->next) {
783 up_write(&uprobe->consumer_rwsem);
788 static int __copy_insn(struct address_space *mapping, struct file *filp,
789 void *insn, int nbytes, loff_t offset)
793 * Ensure that the page that has the original instruction is populated
794 * and in page-cache. If ->readpage == NULL it must be shmem_mapping(),
795 * see uprobe_register().
797 if (mapping->a_ops->readpage)
798 page = read_mapping_page(mapping, offset >> PAGE_SHIFT, filp);
800 page = shmem_read_mapping_page(mapping, offset >> PAGE_SHIFT);
802 return PTR_ERR(page);
804 copy_from_page(page, offset, insn, nbytes);
810 static int copy_insn(struct uprobe *uprobe, struct file *filp)
812 struct address_space *mapping = uprobe->inode->i_mapping;
813 loff_t offs = uprobe->offset;
814 void *insn = &uprobe->arch.insn;
815 int size = sizeof(uprobe->arch.insn);
818 /* Copy only available bytes, -EIO if nothing was read */
820 if (offs >= i_size_read(uprobe->inode))
823 len = min_t(int, size, PAGE_SIZE - (offs & ~PAGE_MASK));
824 err = __copy_insn(mapping, filp, insn, len, offs);
836 static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
837 struct mm_struct *mm, unsigned long vaddr)
841 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
844 /* TODO: move this into _register, until then we abuse this sem. */
845 down_write(&uprobe->consumer_rwsem);
846 if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
849 ret = copy_insn(uprobe, file);
854 if (is_trap_insn((uprobe_opcode_t *)&uprobe->arch.insn))
857 ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
861 smp_wmb(); /* pairs with the smp_rmb() in handle_swbp() */
862 set_bit(UPROBE_COPY_INSN, &uprobe->flags);
865 up_write(&uprobe->consumer_rwsem);
870 static inline bool consumer_filter(struct uprobe_consumer *uc,
871 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
873 return !uc->filter || uc->filter(uc, ctx, mm);
876 static bool filter_chain(struct uprobe *uprobe,
877 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
879 struct uprobe_consumer *uc;
882 down_read(&uprobe->consumer_rwsem);
883 for (uc = uprobe->consumers; uc; uc = uc->next) {
884 ret = consumer_filter(uc, ctx, mm);
888 up_read(&uprobe->consumer_rwsem);
894 install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
895 struct vm_area_struct *vma, unsigned long vaddr)
900 ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr);
905 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
906 * the task can hit this breakpoint right after __replace_page().
908 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
910 set_bit(MMF_HAS_UPROBES, &mm->flags);
912 ret = set_swbp(&uprobe->arch, mm, vaddr);
914 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
915 else if (first_uprobe)
916 clear_bit(MMF_HAS_UPROBES, &mm->flags);
922 remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
924 set_bit(MMF_RECALC_UPROBES, &mm->flags);
925 return set_orig_insn(&uprobe->arch, mm, vaddr);
928 static inline bool uprobe_is_active(struct uprobe *uprobe)
930 return !RB_EMPTY_NODE(&uprobe->rb_node);
933 * There could be threads that have already hit the breakpoint. They
934 * will recheck the current insn and restart if find_uprobe() fails.
935 * See find_active_uprobe().
937 static void delete_uprobe(struct uprobe *uprobe)
939 if (WARN_ON(!uprobe_is_active(uprobe)))
942 spin_lock(&uprobes_treelock);
943 rb_erase(&uprobe->rb_node, &uprobes_tree);
944 spin_unlock(&uprobes_treelock);
945 RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */
950 struct map_info *next;
951 struct mm_struct *mm;
955 static inline struct map_info *free_map_info(struct map_info *info)
957 struct map_info *next = info->next;
962 static struct map_info *
963 build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
965 unsigned long pgoff = offset >> PAGE_SHIFT;
966 struct vm_area_struct *vma;
967 struct map_info *curr = NULL;
968 struct map_info *prev = NULL;
969 struct map_info *info;
973 i_mmap_lock_read(mapping);
974 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
975 if (!valid_vma(vma, is_register))
978 if (!prev && !more) {
980 * Needs GFP_NOWAIT to avoid i_mmap_rwsem recursion through
981 * reclaim. This is optimistic, no harm done if it fails.
983 prev = kmalloc(sizeof(struct map_info),
984 GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
993 if (!mmget_not_zero(vma->vm_mm))
1001 info->mm = vma->vm_mm;
1002 info->vaddr = offset_to_vaddr(vma, offset);
1004 i_mmap_unlock_read(mapping);
1016 info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
1018 curr = ERR_PTR(-ENOMEM);
1028 prev = free_map_info(prev);
1033 register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)
1035 bool is_register = !!new;
1036 struct map_info *info;
1039 percpu_down_write(&dup_mmap_sem);
1040 info = build_map_info(uprobe->inode->i_mapping,
1041 uprobe->offset, is_register);
1043 err = PTR_ERR(info);
1048 struct mm_struct *mm = info->mm;
1049 struct vm_area_struct *vma;
1051 if (err && is_register)
1054 mmap_write_lock(mm);
1055 vma = find_vma(mm, info->vaddr);
1056 if (!vma || !valid_vma(vma, is_register) ||
1057 file_inode(vma->vm_file) != uprobe->inode)
1060 if (vma->vm_start > info->vaddr ||
1061 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
1065 /* consult only the "caller", new consumer. */
1066 if (consumer_filter(new,
1067 UPROBE_FILTER_REGISTER, mm))
1068 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
1069 } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) {
1070 if (!filter_chain(uprobe,
1071 UPROBE_FILTER_UNREGISTER, mm))
1072 err |= remove_breakpoint(uprobe, mm, info->vaddr);
1076 mmap_write_unlock(mm);
1079 info = free_map_info(info);
1082 percpu_up_write(&dup_mmap_sem);
1087 __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
1091 if (WARN_ON(!consumer_del(uprobe, uc)))
1094 err = register_for_each_vma(uprobe, NULL);
1095 /* TODO : cant unregister? schedule a worker thread */
1096 if (!uprobe->consumers && !err)
1097 delete_uprobe(uprobe);
1101 * uprobe_unregister - unregister an already registered probe.
1102 * @inode: the file in which the probe has to be removed.
1103 * @offset: offset from the start of the file.
1104 * @uc: identify which probe if multiple probes are colocated.
1106 void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
1108 struct uprobe *uprobe;
1110 uprobe = find_uprobe(inode, offset);
1111 if (WARN_ON(!uprobe))
1114 down_write(&uprobe->register_rwsem);
1115 __uprobe_unregister(uprobe, uc);
1116 up_write(&uprobe->register_rwsem);
1119 EXPORT_SYMBOL_GPL(uprobe_unregister);
1122 * __uprobe_register - register a probe
1123 * @inode: the file in which the probe has to be placed.
1124 * @offset: offset from the start of the file.
1125 * @uc: information on howto handle the probe..
1127 * Apart from the access refcount, __uprobe_register() takes a creation
1128 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
1129 * inserted into the rbtree (i.e first consumer for a @inode:@offset
1130 * tuple). Creation refcount stops uprobe_unregister from freeing the
1131 * @uprobe even before the register operation is complete. Creation
1132 * refcount is released when the last @uc for the @uprobe
1133 * unregisters. Caller of __uprobe_register() is required to keep @inode
1134 * (and the containing mount) referenced.
1136 * Return errno if it cannot successully install probes
1137 * else return 0 (success)
1139 static int __uprobe_register(struct inode *inode, loff_t offset,
1140 loff_t ref_ctr_offset, struct uprobe_consumer *uc)
1142 struct uprobe *uprobe;
1145 /* Uprobe must have at least one set consumer */
1146 if (!uc->handler && !uc->ret_handler)
1149 /* copy_insn() uses read_mapping_page() or shmem_read_mapping_page() */
1150 if (!inode->i_mapping->a_ops->readpage && !shmem_mapping(inode->i_mapping))
1152 /* Racy, just to catch the obvious mistakes */
1153 if (offset > i_size_read(inode))
1157 * This ensures that copy_from_page(), copy_to_page() and
1158 * __update_ref_ctr() can't cross page boundary.
1160 if (!IS_ALIGNED(offset, UPROBE_SWBP_INSN_SIZE))
1162 if (!IS_ALIGNED(ref_ctr_offset, sizeof(short)))
1166 uprobe = alloc_uprobe(inode, offset, ref_ctr_offset);
1170 return PTR_ERR(uprobe);
1173 * We can race with uprobe_unregister()->delete_uprobe().
1174 * Check uprobe_is_active() and retry if it is false.
1176 down_write(&uprobe->register_rwsem);
1178 if (likely(uprobe_is_active(uprobe))) {
1179 consumer_add(uprobe, uc);
1180 ret = register_for_each_vma(uprobe, uc);
1182 __uprobe_unregister(uprobe, uc);
1184 up_write(&uprobe->register_rwsem);
1187 if (unlikely(ret == -EAGAIN))
1192 int uprobe_register(struct inode *inode, loff_t offset,
1193 struct uprobe_consumer *uc)
1195 return __uprobe_register(inode, offset, 0, uc);
1197 EXPORT_SYMBOL_GPL(uprobe_register);
1199 int uprobe_register_refctr(struct inode *inode, loff_t offset,
1200 loff_t ref_ctr_offset, struct uprobe_consumer *uc)
1202 return __uprobe_register(inode, offset, ref_ctr_offset, uc);
1204 EXPORT_SYMBOL_GPL(uprobe_register_refctr);
1207 * uprobe_apply - unregister an already registered probe.
1208 * @inode: the file in which the probe has to be removed.
1209 * @offset: offset from the start of the file.
1210 * @uc: consumer which wants to add more or remove some breakpoints
1211 * @add: add or remove the breakpoints
1213 int uprobe_apply(struct inode *inode, loff_t offset,
1214 struct uprobe_consumer *uc, bool add)
1216 struct uprobe *uprobe;
1217 struct uprobe_consumer *con;
1220 uprobe = find_uprobe(inode, offset);
1221 if (WARN_ON(!uprobe))
1224 down_write(&uprobe->register_rwsem);
1225 for (con = uprobe->consumers; con && con != uc ; con = con->next)
1228 ret = register_for_each_vma(uprobe, add ? uc : NULL);
1229 up_write(&uprobe->register_rwsem);
1235 static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm)
1237 struct vm_area_struct *vma;
1241 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1242 unsigned long vaddr;
1245 if (!valid_vma(vma, false) ||
1246 file_inode(vma->vm_file) != uprobe->inode)
1249 offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
1250 if (uprobe->offset < offset ||
1251 uprobe->offset >= offset + vma->vm_end - vma->vm_start)
1254 vaddr = offset_to_vaddr(vma, uprobe->offset);
1255 err |= remove_breakpoint(uprobe, mm, vaddr);
1257 mmap_read_unlock(mm);
1262 static struct rb_node *
1263 find_node_in_range(struct inode *inode, loff_t min, loff_t max)
1265 struct rb_node *n = uprobes_tree.rb_node;
1268 struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
1270 if (inode < u->inode) {
1272 } else if (inode > u->inode) {
1275 if (max < u->offset)
1277 else if (min > u->offset)
1288 * For a given range in vma, build a list of probes that need to be inserted.
1290 static void build_probe_list(struct inode *inode,
1291 struct vm_area_struct *vma,
1292 unsigned long start, unsigned long end,
1293 struct list_head *head)
1296 struct rb_node *n, *t;
1299 INIT_LIST_HEAD(head);
1300 min = vaddr_to_offset(vma, start);
1301 max = min + (end - start) - 1;
1303 spin_lock(&uprobes_treelock);
1304 n = find_node_in_range(inode, min, max);
1306 for (t = n; t; t = rb_prev(t)) {
1307 u = rb_entry(t, struct uprobe, rb_node);
1308 if (u->inode != inode || u->offset < min)
1310 list_add(&u->pending_list, head);
1313 for (t = n; (t = rb_next(t)); ) {
1314 u = rb_entry(t, struct uprobe, rb_node);
1315 if (u->inode != inode || u->offset > max)
1317 list_add(&u->pending_list, head);
1321 spin_unlock(&uprobes_treelock);
1324 /* @vma contains reference counter, not the probed instruction. */
1325 static int delayed_ref_ctr_inc(struct vm_area_struct *vma)
1327 struct list_head *pos, *q;
1328 struct delayed_uprobe *du;
1329 unsigned long vaddr;
1330 int ret = 0, err = 0;
1332 mutex_lock(&delayed_uprobe_lock);
1333 list_for_each_safe(pos, q, &delayed_uprobe_list) {
1334 du = list_entry(pos, struct delayed_uprobe, list);
1336 if (du->mm != vma->vm_mm ||
1337 !valid_ref_ctr_vma(du->uprobe, vma))
1340 vaddr = offset_to_vaddr(vma, du->uprobe->ref_ctr_offset);
1341 ret = __update_ref_ctr(vma->vm_mm, vaddr, 1);
1343 update_ref_ctr_warn(du->uprobe, vma->vm_mm, 1);
1347 delayed_uprobe_delete(du);
1349 mutex_unlock(&delayed_uprobe_lock);
1354 * Called from mmap_region/vma_adjust with mm->mmap_lock acquired.
1356 * Currently we ignore all errors and always return 0, the callers
1357 * can't handle the failure anyway.
1359 int uprobe_mmap(struct vm_area_struct *vma)
1361 struct list_head tmp_list;
1362 struct uprobe *uprobe, *u;
1363 struct inode *inode;
1365 if (no_uprobe_events())
1369 (vma->vm_flags & (VM_WRITE|VM_SHARED)) == VM_WRITE &&
1370 test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags))
1371 delayed_ref_ctr_inc(vma);
1373 if (!valid_vma(vma, true))
1376 inode = file_inode(vma->vm_file);
1380 mutex_lock(uprobes_mmap_hash(inode));
1381 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1383 * We can race with uprobe_unregister(), this uprobe can be already
1384 * removed. But in this case filter_chain() must return false, all
1385 * consumers have gone away.
1387 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1388 if (!fatal_signal_pending(current) &&
1389 filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) {
1390 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1391 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1395 mutex_unlock(uprobes_mmap_hash(inode));
1401 vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1404 struct inode *inode;
1407 inode = file_inode(vma->vm_file);
1409 min = vaddr_to_offset(vma, start);
1410 max = min + (end - start) - 1;
1412 spin_lock(&uprobes_treelock);
1413 n = find_node_in_range(inode, min, max);
1414 spin_unlock(&uprobes_treelock);
1420 * Called in context of a munmap of a vma.
1422 void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1424 if (no_uprobe_events() || !valid_vma(vma, false))
1427 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1430 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1431 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1434 if (vma_has_uprobes(vma, start, end))
1435 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1438 /* Slot allocation for XOL */
1439 static int xol_add_vma(struct mm_struct *mm, struct xol_area *area)
1441 struct vm_area_struct *vma;
1444 if (mmap_write_lock_killable(mm))
1447 if (mm->uprobes_state.xol_area) {
1453 /* Try to map as high as possible, this is only a hint. */
1454 area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE,
1456 if (IS_ERR_VALUE(area->vaddr)) {
1462 vma = _install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1463 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO,
1464 &area->xol_mapping);
1471 /* pairs with get_xol_area() */
1472 smp_store_release(&mm->uprobes_state.xol_area, area); /* ^^^ */
1474 mmap_write_unlock(mm);
1479 static struct xol_area *__create_xol_area(unsigned long vaddr)
1481 struct mm_struct *mm = current->mm;
1482 uprobe_opcode_t insn = UPROBE_SWBP_INSN;
1483 struct xol_area *area;
1485 area = kmalloc(sizeof(*area), GFP_KERNEL);
1486 if (unlikely(!area))
1489 area->bitmap = kcalloc(BITS_TO_LONGS(UINSNS_PER_PAGE), sizeof(long),
1494 area->xol_mapping.name = "[uprobes]";
1495 area->xol_mapping.fault = NULL;
1496 area->xol_mapping.pages = area->pages;
1497 area->pages[0] = alloc_page(GFP_HIGHUSER);
1498 if (!area->pages[0])
1500 area->pages[1] = NULL;
1502 area->vaddr = vaddr;
1503 init_waitqueue_head(&area->wq);
1504 /* Reserve the 1st slot for get_trampoline_vaddr() */
1505 set_bit(0, area->bitmap);
1506 atomic_set(&area->slot_count, 1);
1507 arch_uprobe_copy_ixol(area->pages[0], 0, &insn, UPROBE_SWBP_INSN_SIZE);
1509 if (!xol_add_vma(mm, area))
1512 __free_page(area->pages[0]);
1514 kfree(area->bitmap);
1522 * get_xol_area - Allocate process's xol_area if necessary.
1523 * This area will be used for storing instructions for execution out of line.
1525 * Returns the allocated area or NULL.
1527 static struct xol_area *get_xol_area(void)
1529 struct mm_struct *mm = current->mm;
1530 struct xol_area *area;
1532 if (!mm->uprobes_state.xol_area)
1533 __create_xol_area(0);
1535 /* Pairs with xol_add_vma() smp_store_release() */
1536 area = READ_ONCE(mm->uprobes_state.xol_area); /* ^^^ */
1541 * uprobe_clear_state - Free the area allocated for slots.
1543 void uprobe_clear_state(struct mm_struct *mm)
1545 struct xol_area *area = mm->uprobes_state.xol_area;
1547 mutex_lock(&delayed_uprobe_lock);
1548 delayed_uprobe_remove(NULL, mm);
1549 mutex_unlock(&delayed_uprobe_lock);
1554 put_page(area->pages[0]);
1555 kfree(area->bitmap);
1559 void uprobe_start_dup_mmap(void)
1561 percpu_down_read(&dup_mmap_sem);
1564 void uprobe_end_dup_mmap(void)
1566 percpu_up_read(&dup_mmap_sem);
1569 void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1571 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1572 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1573 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1574 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1579 * - search for a free slot.
1581 static unsigned long xol_take_insn_slot(struct xol_area *area)
1583 unsigned long slot_addr;
1587 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1588 if (slot_nr < UINSNS_PER_PAGE) {
1589 if (!test_and_set_bit(slot_nr, area->bitmap))
1592 slot_nr = UINSNS_PER_PAGE;
1595 wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1596 } while (slot_nr >= UINSNS_PER_PAGE);
1598 slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1599 atomic_inc(&area->slot_count);
1605 * xol_get_insn_slot - allocate a slot for xol.
1606 * Returns the allocated slot address or 0.
1608 static unsigned long xol_get_insn_slot(struct uprobe *uprobe)
1610 struct xol_area *area;
1611 unsigned long xol_vaddr;
1613 area = get_xol_area();
1617 xol_vaddr = xol_take_insn_slot(area);
1618 if (unlikely(!xol_vaddr))
1621 arch_uprobe_copy_ixol(area->pages[0], xol_vaddr,
1622 &uprobe->arch.ixol, sizeof(uprobe->arch.ixol));
1628 * xol_free_insn_slot - If slot was earlier allocated by
1629 * @xol_get_insn_slot(), make the slot available for
1630 * subsequent requests.
1632 static void xol_free_insn_slot(struct task_struct *tsk)
1634 struct xol_area *area;
1635 unsigned long vma_end;
1636 unsigned long slot_addr;
1638 if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1641 slot_addr = tsk->utask->xol_vaddr;
1642 if (unlikely(!slot_addr))
1645 area = tsk->mm->uprobes_state.xol_area;
1646 vma_end = area->vaddr + PAGE_SIZE;
1647 if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1648 unsigned long offset;
1651 offset = slot_addr - area->vaddr;
1652 slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1653 if (slot_nr >= UINSNS_PER_PAGE)
1656 clear_bit(slot_nr, area->bitmap);
1657 atomic_dec(&area->slot_count);
1658 smp_mb__after_atomic(); /* pairs with prepare_to_wait() */
1659 if (waitqueue_active(&area->wq))
1662 tsk->utask->xol_vaddr = 0;
1666 void __weak arch_uprobe_copy_ixol(struct page *page, unsigned long vaddr,
1667 void *src, unsigned long len)
1669 /* Initialize the slot */
1670 copy_to_page(page, vaddr, src, len);
1673 * We probably need flush_icache_user_page() but it needs vma.
1674 * This should work on most of architectures by default. If
1675 * architecture needs to do something different it can define
1676 * its own version of the function.
1678 flush_dcache_page(page);
1682 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1683 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1685 * Return the address of the breakpoint instruction.
1687 unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1689 return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1692 unsigned long uprobe_get_trap_addr(struct pt_regs *regs)
1694 struct uprobe_task *utask = current->utask;
1696 if (unlikely(utask && utask->active_uprobe))
1697 return utask->vaddr;
1699 return instruction_pointer(regs);
1702 static struct return_instance *free_ret_instance(struct return_instance *ri)
1704 struct return_instance *next = ri->next;
1705 put_uprobe(ri->uprobe);
1711 * Called with no locks held.
1712 * Called in context of an exiting or an exec-ing thread.
1714 void uprobe_free_utask(struct task_struct *t)
1716 struct uprobe_task *utask = t->utask;
1717 struct return_instance *ri;
1722 if (utask->active_uprobe)
1723 put_uprobe(utask->active_uprobe);
1725 ri = utask->return_instances;
1727 ri = free_ret_instance(ri);
1729 xol_free_insn_slot(t);
1735 * Allocate a uprobe_task object for the task if necessary.
1736 * Called when the thread hits a breakpoint.
1739 * - pointer to new uprobe_task on success
1742 static struct uprobe_task *get_utask(void)
1744 if (!current->utask)
1745 current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1746 return current->utask;
1749 static int dup_utask(struct task_struct *t, struct uprobe_task *o_utask)
1751 struct uprobe_task *n_utask;
1752 struct return_instance **p, *o, *n;
1754 n_utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL);
1759 p = &n_utask->return_instances;
1760 for (o = o_utask->return_instances; o; o = o->next) {
1761 n = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1766 get_uprobe(n->uprobe);
1777 static void uprobe_warn(struct task_struct *t, const char *msg)
1779 pr_warn("uprobe: %s:%d failed to %s\n",
1780 current->comm, current->pid, msg);
1783 static void dup_xol_work(struct callback_head *work)
1785 if (current->flags & PF_EXITING)
1788 if (!__create_xol_area(current->utask->dup_xol_addr) &&
1789 !fatal_signal_pending(current))
1790 uprobe_warn(current, "dup xol area");
1794 * Called in context of a new clone/fork from copy_process.
1796 void uprobe_copy_process(struct task_struct *t, unsigned long flags)
1798 struct uprobe_task *utask = current->utask;
1799 struct mm_struct *mm = current->mm;
1800 struct xol_area *area;
1804 if (!utask || !utask->return_instances)
1807 if (mm == t->mm && !(flags & CLONE_VFORK))
1810 if (dup_utask(t, utask))
1811 return uprobe_warn(t, "dup ret instances");
1813 /* The task can fork() after dup_xol_work() fails */
1814 area = mm->uprobes_state.xol_area;
1816 return uprobe_warn(t, "dup xol area");
1821 t->utask->dup_xol_addr = area->vaddr;
1822 init_task_work(&t->utask->dup_xol_work, dup_xol_work);
1823 task_work_add(t, &t->utask->dup_xol_work, TWA_RESUME);
1827 * Current area->vaddr notion assume the trampoline address is always
1828 * equal area->vaddr.
1830 * Returns -1 in case the xol_area is not allocated.
1832 static unsigned long get_trampoline_vaddr(void)
1834 struct xol_area *area;
1835 unsigned long trampoline_vaddr = -1;
1837 /* Pairs with xol_add_vma() smp_store_release() */
1838 area = READ_ONCE(current->mm->uprobes_state.xol_area); /* ^^^ */
1840 trampoline_vaddr = area->vaddr;
1842 return trampoline_vaddr;
1845 static void cleanup_return_instances(struct uprobe_task *utask, bool chained,
1846 struct pt_regs *regs)
1848 struct return_instance *ri = utask->return_instances;
1849 enum rp_check ctx = chained ? RP_CHECK_CHAIN_CALL : RP_CHECK_CALL;
1851 while (ri && !arch_uretprobe_is_alive(ri, ctx, regs)) {
1852 ri = free_ret_instance(ri);
1855 utask->return_instances = ri;
1858 static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs)
1860 struct return_instance *ri;
1861 struct uprobe_task *utask;
1862 unsigned long orig_ret_vaddr, trampoline_vaddr;
1865 if (!get_xol_area())
1868 utask = get_utask();
1872 if (utask->depth >= MAX_URETPROBE_DEPTH) {
1873 printk_ratelimited(KERN_INFO "uprobe: omit uretprobe due to"
1874 " nestedness limit pid/tgid=%d/%d\n",
1875 current->pid, current->tgid);
1879 ri = kmalloc(sizeof(struct return_instance), GFP_KERNEL);
1883 trampoline_vaddr = get_trampoline_vaddr();
1884 orig_ret_vaddr = arch_uretprobe_hijack_return_addr(trampoline_vaddr, regs);
1885 if (orig_ret_vaddr == -1)
1888 /* drop the entries invalidated by longjmp() */
1889 chained = (orig_ret_vaddr == trampoline_vaddr);
1890 cleanup_return_instances(utask, chained, regs);
1893 * We don't want to keep trampoline address in stack, rather keep the
1894 * original return address of first caller thru all the consequent
1895 * instances. This also makes breakpoint unwrapping easier.
1898 if (!utask->return_instances) {
1900 * This situation is not possible. Likely we have an
1901 * attack from user-space.
1903 uprobe_warn(current, "handle tail call");
1906 orig_ret_vaddr = utask->return_instances->orig_ret_vaddr;
1909 ri->uprobe = get_uprobe(uprobe);
1910 ri->func = instruction_pointer(regs);
1911 ri->stack = user_stack_pointer(regs);
1912 ri->orig_ret_vaddr = orig_ret_vaddr;
1913 ri->chained = chained;
1916 ri->next = utask->return_instances;
1917 utask->return_instances = ri;
1924 /* Prepare to single-step probed instruction out of line. */
1926 pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr)
1928 struct uprobe_task *utask;
1929 unsigned long xol_vaddr;
1932 utask = get_utask();
1936 xol_vaddr = xol_get_insn_slot(uprobe);
1940 utask->xol_vaddr = xol_vaddr;
1941 utask->vaddr = bp_vaddr;
1943 err = arch_uprobe_pre_xol(&uprobe->arch, regs);
1944 if (unlikely(err)) {
1945 xol_free_insn_slot(current);
1949 utask->active_uprobe = uprobe;
1950 utask->state = UTASK_SSTEP;
1955 * If we are singlestepping, then ensure this thread is not connected to
1956 * non-fatal signals until completion of singlestep. When xol insn itself
1957 * triggers the signal, restart the original insn even if the task is
1958 * already SIGKILL'ed (since coredump should report the correct ip). This
1959 * is even more important if the task has a handler for SIGSEGV/etc, The
1960 * _same_ instruction should be repeated again after return from the signal
1961 * handler, and SSTEP can never finish in this case.
1963 bool uprobe_deny_signal(void)
1965 struct task_struct *t = current;
1966 struct uprobe_task *utask = t->utask;
1968 if (likely(!utask || !utask->active_uprobe))
1971 WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1973 if (task_sigpending(t)) {
1974 spin_lock_irq(&t->sighand->siglock);
1975 clear_tsk_thread_flag(t, TIF_SIGPENDING);
1976 spin_unlock_irq(&t->sighand->siglock);
1978 if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1979 utask->state = UTASK_SSTEP_TRAPPED;
1980 set_tsk_thread_flag(t, TIF_UPROBE);
1987 static void mmf_recalc_uprobes(struct mm_struct *mm)
1989 struct vm_area_struct *vma;
1991 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1992 if (!valid_vma(vma, false))
1995 * This is not strictly accurate, we can race with
1996 * uprobe_unregister() and see the already removed
1997 * uprobe if delete_uprobe() was not yet called.
1998 * Or this uprobe can be filtered out.
2000 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
2004 clear_bit(MMF_HAS_UPROBES, &mm->flags);
2007 static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr)
2010 uprobe_opcode_t opcode;
2013 if (WARN_ON_ONCE(!IS_ALIGNED(vaddr, UPROBE_SWBP_INSN_SIZE)))
2016 pagefault_disable();
2017 result = __get_user(opcode, (uprobe_opcode_t __user *)vaddr);
2020 if (likely(result == 0))
2024 * The NULL 'tsk' here ensures that any faults that occur here
2025 * will not be accounted to the task. 'mm' *is* current->mm,
2026 * but we treat this as a 'remote' access since it is
2027 * essentially a kernel access to the memory.
2029 result = get_user_pages_remote(mm, vaddr, 1, FOLL_FORCE, &page,
2034 copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
2037 /* This needs to return true for any variant of the trap insn */
2038 return is_trap_insn(&opcode);
2041 static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
2043 struct mm_struct *mm = current->mm;
2044 struct uprobe *uprobe = NULL;
2045 struct vm_area_struct *vma;
2048 vma = vma_lookup(mm, bp_vaddr);
2050 if (valid_vma(vma, false)) {
2051 struct inode *inode = file_inode(vma->vm_file);
2052 loff_t offset = vaddr_to_offset(vma, bp_vaddr);
2054 uprobe = find_uprobe(inode, offset);
2058 *is_swbp = is_trap_at_addr(mm, bp_vaddr);
2063 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
2064 mmf_recalc_uprobes(mm);
2065 mmap_read_unlock(mm);
2070 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
2072 struct uprobe_consumer *uc;
2073 int remove = UPROBE_HANDLER_REMOVE;
2074 bool need_prep = false; /* prepare return uprobe, when needed */
2076 down_read(&uprobe->register_rwsem);
2077 for (uc = uprobe->consumers; uc; uc = uc->next) {
2081 rc = uc->handler(uc, regs);
2082 WARN(rc & ~UPROBE_HANDLER_MASK,
2083 "bad rc=0x%x from %ps()\n", rc, uc->handler);
2086 if (uc->ret_handler)
2092 if (need_prep && !remove)
2093 prepare_uretprobe(uprobe, regs); /* put bp at return */
2095 if (remove && uprobe->consumers) {
2096 WARN_ON(!uprobe_is_active(uprobe));
2097 unapply_uprobe(uprobe, current->mm);
2099 up_read(&uprobe->register_rwsem);
2103 handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs)
2105 struct uprobe *uprobe = ri->uprobe;
2106 struct uprobe_consumer *uc;
2108 down_read(&uprobe->register_rwsem);
2109 for (uc = uprobe->consumers; uc; uc = uc->next) {
2110 if (uc->ret_handler)
2111 uc->ret_handler(uc, ri->func, regs);
2113 up_read(&uprobe->register_rwsem);
2116 static struct return_instance *find_next_ret_chain(struct return_instance *ri)
2121 chained = ri->chained;
2122 ri = ri->next; /* can't be NULL if chained */
2128 static void handle_trampoline(struct pt_regs *regs)
2130 struct uprobe_task *utask;
2131 struct return_instance *ri, *next;
2134 utask = current->utask;
2138 ri = utask->return_instances;
2144 * We should throw out the frames invalidated by longjmp().
2145 * If this chain is valid, then the next one should be alive
2146 * or NULL; the latter case means that nobody but ri->func
2147 * could hit this trampoline on return. TODO: sigaltstack().
2149 next = find_next_ret_chain(ri);
2150 valid = !next || arch_uretprobe_is_alive(next, RP_CHECK_RET, regs);
2152 instruction_pointer_set(regs, ri->orig_ret_vaddr);
2155 handle_uretprobe_chain(ri, regs);
2156 ri = free_ret_instance(ri);
2158 } while (ri != next);
2161 utask->return_instances = ri;
2165 uprobe_warn(current, "handle uretprobe, sending SIGILL.");
2170 bool __weak arch_uprobe_ignore(struct arch_uprobe *aup, struct pt_regs *regs)
2175 bool __weak arch_uretprobe_is_alive(struct return_instance *ret, enum rp_check ctx,
2176 struct pt_regs *regs)
2182 * Run handler and ask thread to singlestep.
2183 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
2185 static void handle_swbp(struct pt_regs *regs)
2187 struct uprobe *uprobe;
2188 unsigned long bp_vaddr;
2191 bp_vaddr = uprobe_get_swbp_addr(regs);
2192 if (bp_vaddr == get_trampoline_vaddr())
2193 return handle_trampoline(regs);
2195 uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
2198 /* No matching uprobe; signal SIGTRAP. */
2202 * Either we raced with uprobe_unregister() or we can't
2203 * access this memory. The latter is only possible if
2204 * another thread plays with our ->mm. In both cases
2205 * we can simply restart. If this vma was unmapped we
2206 * can pretend this insn was not executed yet and get
2207 * the (correct) SIGSEGV after restart.
2209 instruction_pointer_set(regs, bp_vaddr);
2214 /* change it in advance for ->handler() and restart */
2215 instruction_pointer_set(regs, bp_vaddr);
2218 * TODO: move copy_insn/etc into _register and remove this hack.
2219 * After we hit the bp, _unregister + _register can install the
2220 * new and not-yet-analyzed uprobe at the same address, restart.
2222 if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
2226 * Pairs with the smp_wmb() in prepare_uprobe().
2228 * Guarantees that if we see the UPROBE_COPY_INSN bit set, then
2229 * we must also see the stores to &uprobe->arch performed by the
2230 * prepare_uprobe() call.
2234 /* Tracing handlers use ->utask to communicate with fetch methods */
2238 if (arch_uprobe_ignore(&uprobe->arch, regs))
2241 handler_chain(uprobe, regs);
2243 if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
2246 if (!pre_ssout(uprobe, regs, bp_vaddr))
2249 /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */
2255 * Perform required fix-ups and disable singlestep.
2256 * Allow pending signals to take effect.
2258 static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
2260 struct uprobe *uprobe;
2263 uprobe = utask->active_uprobe;
2264 if (utask->state == UTASK_SSTEP_ACK)
2265 err = arch_uprobe_post_xol(&uprobe->arch, regs);
2266 else if (utask->state == UTASK_SSTEP_TRAPPED)
2267 arch_uprobe_abort_xol(&uprobe->arch, regs);
2272 utask->active_uprobe = NULL;
2273 utask->state = UTASK_RUNNING;
2274 xol_free_insn_slot(current);
2276 spin_lock_irq(¤t->sighand->siglock);
2277 recalc_sigpending(); /* see uprobe_deny_signal() */
2278 spin_unlock_irq(¤t->sighand->siglock);
2280 if (unlikely(err)) {
2281 uprobe_warn(current, "execute the probed insn, sending SIGILL.");
2287 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and
2288 * allows the thread to return from interrupt. After that handle_swbp()
2289 * sets utask->active_uprobe.
2291 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag
2292 * and allows the thread to return from interrupt.
2294 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
2295 * uprobe_notify_resume().
2297 void uprobe_notify_resume(struct pt_regs *regs)
2299 struct uprobe_task *utask;
2301 clear_thread_flag(TIF_UPROBE);
2303 utask = current->utask;
2304 if (utask && utask->active_uprobe)
2305 handle_singlestep(utask, regs);
2311 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
2312 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
2314 int uprobe_pre_sstep_notifier(struct pt_regs *regs)
2319 if (!test_bit(MMF_HAS_UPROBES, ¤t->mm->flags) &&
2320 (!current->utask || !current->utask->return_instances))
2323 set_thread_flag(TIF_UPROBE);
2328 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
2329 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
2331 int uprobe_post_sstep_notifier(struct pt_regs *regs)
2333 struct uprobe_task *utask = current->utask;
2335 if (!current->mm || !utask || !utask->active_uprobe)
2336 /* task is currently not uprobed */
2339 utask->state = UTASK_SSTEP_ACK;
2340 set_thread_flag(TIF_UPROBE);
2344 static struct notifier_block uprobe_exception_nb = {
2345 .notifier_call = arch_uprobe_exception_notify,
2346 .priority = INT_MAX-1, /* notified after kprobes, kgdb */
2349 void __init uprobes_init(void)
2353 for (i = 0; i < UPROBES_HASH_SZ; i++)
2354 mutex_init(&uprobes_mmap_mutex[i]);
2356 BUG_ON(register_die_notifier(&uprobe_exception_nb));